Non-stationarity analysis of flood flows using copula based change-point detection method: Application to case study...

This study investigates the change in dependence structure of bivariate flood flow characteristics namely magnitude, frequency and timing in the Godavari river basin using a copula based likelihood ratio (CLR) test. Parametric bootstrap was used to obtain critical values of CLR test for the best-fitted copula. The performance of the CLR method was also evaluated for simulated synthetic bivariate series with a known change-point location in the dependence (copula parameter). Then the methodology was applied to two streamflow monitoring sites Bhimkund and Wagholi-Butti, located in the Godavari river basin in India. Streamflow data for 33 years from 1977 to 2009 was analyzed, by extracting the series of flow characteristics of magnitude, frequency and timing. Initially univariate change-point detection (CPD) test namely standard normal homogeneity test was applied to detect abrupt change-point in mean of the flow series. At Bhimkund site, there was abrupt increase in mean of magnitude, frequency and timing series after the identified change-point year. However, at Wagoli-Butti site, there was abrupt decrease in mean of magnitude and frequency series although timing series got delayed (i.e., abrupt increase). After univariate CPD in mean, the bivariate series i.e., magnitude-frequency and magnitude-timing pairs for these sites were analyzed to detect the change-points in dependence in terms of copula parameters using the CLR method. The results showed that change-points in the copula parameters were detected at year 2003 and 2004 for Wagoli- Butti and Bhimkund sites respectively, and appear to be jointly non-stationary due to human induced change at these two sites. The results of study for detection of the change-point location in the dependency structure of flow characteristics would be useful for flood risk assessment in the basin.